Recognition and Classification of Allergy-Related Events from Wearable Sensor Data

• Review methods for event recognition and time-series classification
• Preprocess, synchronize, and segment multimodal wearable-sensor data
• Develop a reproducible pipeline for data preparation, model training, validation, and testing
• Train an event-recognition model to distinguish allergy-related events from background activities
• Train a classification model to differentiate between event types such as eye rubbing, palate rubbing, swallowing, and throat clearing
• Compare classical machine-learning and deep-learning approaches
• Investigate suitable temporal windowing and event-segmentation strategies
• Explore multimodal sensor fusion and assess the contribution of individual sensor modalities
• Address challenges such as class imbalance, sensor noise, motion artifacts, and inter-person variability
• Evaluate the pipeline using event-based, classification, and participant-independent metrics
• Analyze false detections and confusion between similar event classes

The scope will be adapted to the requirements of a Bachelor’s or Master’s thesis. A Bachelor’s thesis may focus on implementing and systematically comparing established recognition and classification approaches, while a Master’s thesis may investigate advanced multimodal architectures, personalization, or generalization to previously unseen users and realistic everyday conditions.

You should:

• Study computer science, data science, electrical engineering, medical engineering, or a related discipline
• Have practical experience developing a machine-learning pipeline and training models independently
• Have good Python programming skills
• Be familiar with PyTorch, TensorFlow, scikit-learn, or comparable frameworks
• Understand model evaluation, cross-validation, and the prevention of data leakage
• Be interested in wearable sensing, time-series analysis, or human activity recognition
• Be able to work independently and systematically
• Be fluent in English or German

It would be great if you:

• Have experience with multivariate or multimodal time-series data
• Have worked with IMU, acoustic, physiological, or other wearable-sensor signals
• Are familiar with temporal convolutional networks, recurrent neural networks, transformers, or similar architectures
• Have experience with imbalanced datasets, event-based evaluation, or subject-independent validation
• Are interested in personalization, transfer learning, or uncertainty estimation